Dear statalist,
I finished a manuscript and the regression model (simplified) I used in my analysis is
𝑦 ~ 𝑥1+𝑥2+𝑥2^2+𝑥1:𝑥2+𝑥1:𝑥2^2.
where x1 and x2 are both continuous variable.
I interpreted the interaction results in the original manuscript using the simple slope method. For example, I calculated the conditional effect of x1 on multiple pre-specified values of x2 and compared whether any of those conditional effects are significantly different from zero and whether there is a significant difference in any of those conditional effects by looking at the confidence interval of each conditional effect.
However, after I sent this to all my co-authors, many thought simple slope was not the correct way to interpret interaction effects. They asked me to just put the original regression outputs in the manuscript and explain the parameters I obtained in the model.
My question is since my model includes both linear and quadratic interaction terms, how can I interpret and test if the overall interaction effects are significant in my model? Based on the toy example below, I have two options
First, report the p-value of linear (x1*x2) and the quadratic interaction term (x1*x2^2) separately. But since the quadratic interaction term is non-significant, how can I interpret this? And how can I interpret the overall interaction effects?
Second, using other statistics like the likelihood ratio test (LRT) to compare the full model (with both quadratic and linear interaction terms) and nested model (without any interaction terms) and report the LRT statistics.
Which method is correct in interpreting the significance level in my question? And do you think the simple slope (conditional) is sufficient in my example to interpret the results?
I finished a manuscript and the regression model (simplified) I used in my analysis is
𝑦 ~ 𝑥1+𝑥2+𝑥2^2+𝑥1:𝑥2+𝑥1:𝑥2^2.
where x1 and x2 are both continuous variable.
I interpreted the interaction results in the original manuscript using the simple slope method. For example, I calculated the conditional effect of x1 on multiple pre-specified values of x2 and compared whether any of those conditional effects are significantly different from zero and whether there is a significant difference in any of those conditional effects by looking at the confidence interval of each conditional effect.
However, after I sent this to all my co-authors, many thought simple slope was not the correct way to interpret interaction effects. They asked me to just put the original regression outputs in the manuscript and explain the parameters I obtained in the model.
My question is since my model includes both linear and quadratic interaction terms, how can I interpret and test if the overall interaction effects are significant in my model? Based on the toy example below, I have two options
First, report the p-value of linear (x1*x2) and the quadratic interaction term (x1*x2^2) separately. But since the quadratic interaction term is non-significant, how can I interpret this? And how can I interpret the overall interaction effects?
Second, using other statistics like the likelihood ratio test (LRT) to compare the full model (with both quadratic and linear interaction terms) and nested model (without any interaction terms) and report the LRT statistics.
Which method is correct in interpreting the significance level in my question? And do you think the simple slope (conditional) is sufficient in my example to interpret the results?
Attached Files
Comment